Disaster reduction rescue early warning method and system based on artificial intelligence

Abstract

The invention discloses a disaster reduction rescue early warning method and system based on artificial intelligence. The method comprises the steps of data acquisition, disaster diffusion prediction, disaster diffusion range quantification, prediction result evaluation, parameter optimization and disaster reduction rescue early warning. The invention belongs to the field of rescue early warning, and particularly relates to a disaster reduction rescue early warning method and system based on artificial intelligence, and the method comprises the steps: mapping an angle error to a unit arc length through boundary continuous processing, and cooperating with a disaster output normalization layer, thereby preventing a rescue team from propelling towards a disaster source; based on a dynamic error sensitive weighting mechanism, error distribution estimation of dynamic bandwidth is introduced, and it is ensured that the evacuation radius contains a complete risk area; the decision effectiveness is improved based on the risk adaptability of prediction result evaluation; stage regulation and control of disturbance amplitude are introduced, and the optimization convergence speed is increased; the diversity of the current solution is reserved based on the dynamic balance of the core control weight; on the basis of low-efficiency solution replacement, parameter combination population diversity maintenance is ensured; and the rescue early warning effect is improved.

Description

TECHNICAL FIELD

[0001] The application relates to the technical field of rescue early warning, in particular to a disaster reduction and rescue early warning method and system based on artificial intelligence. BACKGROUND

[0002] The disaster reduction and rescue early warning method is a technical means for collecting and analyzing disaster-related data and predicting the disaster diffusion trend by means of a model. The core is to generate early warning information to provide a basis for decision-making such as evacuation planning and rescue deployment to cope with disasters in advance and reduce losses. However, the general disaster reduction and rescue early warning method has the problems of boundary fracture in direction prediction, equalization defects in error processing, static nature of uncertainty quantification, and risk mismatch of evaluation indexes, which leads to poor accuracy of rescue early warning. The general disaster reduction and rescue early warning method has the problems of static iteration process of parameter optimization and degeneration of group diversity, which leads to poor rescue early warning effect. SUMMARY

[0003] In view of the above problems, the application provides a disaster reduction and rescue early warning method and system based on artificial intelligence to overcome the defects of the prior art. The application solves the problems of boundary fracture in direction prediction, equalization defects in error processing, static nature of uncertainty quantification, and risk mismatch of evaluation indexes in the general disaster reduction and rescue early warning method, which leads to poor accuracy of rescue early warning. The application maps the angle error to the unit circular arc length by boundary continuous processing, cooperates with the disaster output normalization layer, and forcibly constrains the results to solve the boundary fracture and avoid the rescue team advancing towards the disaster source. The application introduces error classification weight based on the dynamic error sensitivity weighting mechanism to realize the nonlinear incremental of large error weight. The application introduces the dynamic bandwidth error distribution estimation combined with the cross-border range correction to ensure that the evacuation radius contains the complete risk area. The application improves the decision effectiveness based on the risk adaptability of the prediction result evaluation, and further improves the accuracy of rescue early warning. The application introduces the phased regulation and control of disturbance amplitude to speed up the optimization convergence speed. The application focuses on retaining the diversity of the current solution based on the dynamic balance of the core control weight. The application ensures the maintenance of the diversity of the parameter combination group based on the replacement of inefficient solutions, and further improves the effect of rescue early warning.

[0004] The technical scheme adopted by the application is as follows: The application provides a disaster reduction and rescue early warning method based on artificial intelligence, which comprises the following steps:

[0005] Step S1: data collection;

[0006] Step S2: disaster diffusion prediction;

[0007] Step S3: disaster diffusion range quantification;

[0008] Step S4: Evaluation of prediction results;

[0009] Step S5: Parameter optimization;

[0010] Step S6: Disaster Reduction and Relief Early Warning.

[0011] Further, in step S1, the data acquisition involves obtaining historical disaster spread data; labeling the disaster spread direction; and performing preprocessing to obtain a disaster spread dataset.

[0012] Further, in step S2, the disaster spread prediction is based on a disaster spread dataset and a bidirectional LSTM to establish a disaster spread prediction model to predict the direction of disaster spread. The model structure includes: an input layer, a bidirectional LSTM layer, a fully connected layer, and a disaster output normalization layer; a disaster point loss function is constructed, a dynamic error-sensitive weighting mechanism is introduced, and the correction error of a single sample is defined; error classification weights are designed; the final loss function is obtained; and a disaster output normalization layer is constructed.

[0013] Furthermore, in step S3, the quantification of the disaster spread range is based on the error distribution estimation of the point prediction error, fitting the error distribution, and introducing the dynamic bandwidth of the dynamic time scale; the predicted range obtained by combining the predicted disaster spread direction with the prediction error is processed across the boundary range.

[0014] Further, in step S4, the evaluation of the prediction results involves dividing the disaster spread dataset into a test set and a training set, training the disaster spread prediction model based on the training set, and verifying the performance of the disaster spread prediction model based on the test set; constructing disaster spread prediction evaluation indicators; constructing range prediction evaluation indicators, including prediction range coverage and prediction range normalized average width, and introducing a coverage calibration term; when the disaster spread prediction model converges to the loss on the training set, the disaster spread prediction model training is complete; setting a verification threshold, verifying the model performance based on the verification threshold; if the threshold is met, the disaster spread prediction model is established; otherwise, parameter optimization is performed.

[0015] Further, in step S5, the parameter optimization involves optimizing the parameters of the disaster spread prediction model to minimize the loss function of the validation set, specifically including the following steps:

[0016] Step S51: Initial solution generation; Initial solutions are generated using parameter combination mapping; Each solution represents a set of hyperparameter combinations;

[0017] Step S52: Fitness function design; for each solution, train a bidirectional LSTM and compute the fitness function on the validation set. , as the fitness value;

[0018] Step S53: Adjust the perturbation amplitude; dynamically adjust the perturbation amplitude according to the iteration process; and perform mutation operations;

[0019] Step S54: Inefficient replacement; if the fitness value improvement of the solution after K consecutive iterations is less than the threshold, it is marked as inefficient; replace the inefficient solution with the generated initial solution.

[0020] Step S55: Unlock and update;

[0021] Step S56: Solution evaluation; Set the maximum number of iterations. If the maximum number of iterations is reached or the fitness value of the optimal solution converges, then evaluate the performance of the disaster spread prediction model trained based on the optimal solution. If the standard is met, the disaster spread prediction model is established; otherwise, return to step S51.

[0022] Furthermore, in step S6, the disaster reduction and relief early warning involves collecting disaster spread data in real time, preprocessing it, and then inputting it into the disaster spread prediction model. The obtained disaster spread direction and error distribution are then used to provide early warnings to management personnel.

[0023] The disaster reduction and relief early warning system based on artificial intelligence provided by this invention includes a data acquisition module, a disaster spread prediction module, a disaster spread range quantification module, a prediction result evaluation module, a parameter optimization module, and a disaster reduction and relief early warning module.

[0024] The data acquisition module obtains historical disaster spread data to obtain a disaster spread dataset;

[0025] The disaster spread prediction module is based on a disaster spread dataset and a bidirectional LSTM. It designs a disaster spread prediction model by constructing a loss function that includes a dynamic error-sensitive weighting mechanism.

[0026] The disaster spread range quantification module introduces a dynamic bandwidth fitting error distribution with a dynamic time scale to quantify the disaster spread prediction range and perform cross-boundary range processing.

[0027] The prediction result evaluation module evaluates the prediction performance using the boundary correction average error index to determine whether the disaster spread prediction model meets the standard.

[0028] The parameter optimization module optimizes the parameters of the disaster spread prediction model by dynamically adjusting the disturbance amplitude and core control weights.

[0029] The disaster reduction and relief early warning module is based on a disaster spread prediction model and provides early warning of disaster spread based on real-time collected disaster spread data.

[0030] The beneficial effects achieved by the present invention using the above solution are as follows:

[0031] (1) To address the problems of boundary breakage in directional prediction, equalization defects in error processing, static nature of uncertainty quantification, and risk mismatch in assessment indicators in general disaster reduction and early warning methods, which lead to poor accuracy of rescue early warning, this scheme maps angular errors to unit arc length through boundary continuity processing. Combined with the disaster output normalization layer, it forces the results to resolve boundary breakage and prevent rescue teams from advancing towards the disaster source. Based on the dynamic error-sensitive weighting mechanism, it introduces error classification weights to achieve nonlinear increment of large error weights. It introduces dynamic bandwidth error distribution estimation and combines it with cross-boundary range correction to ensure that the evacuation radius includes the complete risk area. It improves decision-making effectiveness based on the risk adaptability assessment of prediction results, thereby improving the accuracy of rescue early warning.

[0032] (2) In view of the problem that the general disaster reduction and rescue early warning method has a static iterative process for parameter optimization and the degradation of group diversity, which leads to poor rescue early warning effect, this scheme introduces the staged control of the disturbance amplitude to accelerate the optimization convergence speed; the dynamic balance based on the core control weight focuses on retaining the diversity of the current solution; and the replacement of inefficient solutions ensures that the diversity of parameter combination group is maintained; thereby improving the rescue early warning effect. Attached Figure Description

[0033] Figure 1 A flowchart illustrating the artificial intelligence-based disaster reduction and relief early warning method provided by this invention;

[0034] Figure 2 This is a schematic diagram of the artificial intelligence-based disaster reduction and relief early warning system provided by the present invention.

[0035] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used together with the embodiments of the invention to explain the invention and do not constitute a limitation thereof. Detailed Implementation

[0036] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0037] In the description of this invention, it should be understood that the terms "upper", "lower", "front", "rear", "left", "right", "top", "bottom", "inner", "outer", etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are only for the convenience of describing this invention and simplifying the description, and do not indicate or imply that the system or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on this invention.

[0038] Example 1, see Figure 1 The present invention provides an artificial intelligence-based disaster reduction and relief early warning method, which includes the following steps:

[0039] Step S1: Data Acquisition; Obtain historical disaster spread data to obtain a disaster spread dataset;

[0040] Step S2: Disaster spread prediction; Based on the disaster spread dataset and bidirectional LSTM, a disaster spread prediction model is designed by constructing a loss function that includes a dynamic error-sensitive weighting mechanism;

[0041] Step S3: Quantify the disaster spread range; Introduce a dynamic bandwidth fitting error distribution with a dynamic time scale to quantify the disaster spread prediction range and perform cross-boundary range processing;

[0042] Step S4: Evaluation of prediction results; The prediction performance is evaluated by using the boundary correction average error index to determine whether the disaster spread prediction model meets the standards;

[0043] Step S5: Parameter optimization; The parameters of the disaster spread prediction model are optimized by dynamically adjusting the disturbance amplitude and core control weights;

[0044] Step S6: Disaster Reduction and Relief Early Warning; Based on the disaster spread prediction model, provide early warning for disaster relief based on the real-time collected disaster spread data.

[0045] Example 2, see Figure 1 This embodiment is based on the above embodiment. In step S1, data acquisition involves obtaining historical disaster diffusion data. The historical disaster diffusion data includes wind direction, wind speed, air humidity, temperature, air pressure, disaster type, disaster intensity, precipitation, and diffusion source location. The label is the disaster diffusion direction, 0-360 degrees. Preprocessing is performed, including interpolation, filtering, and normalization. The disaster diffusion dataset is then obtained.

[0046] Example 3, see Figure 1This embodiment is based on the above embodiment. In step S2, disaster spread prediction is based on a disaster spread dataset and a bidirectional LSTM to establish a disaster spread prediction model to predict the disaster spread direction in the next 1-3 days. In disaster reduction and relief, it is necessary to predict the disaster spread direction in advance to allow time for evacuation route planning and deployment of rescue forces. It is necessary to solve the problem of boundary discontinuity to avoid misjudging the spread direction. The disaster spread prediction model is constructed, and the temporal features of the disaster spread data are extracted. The model structure includes: an input layer; a bidirectional LSTM layer (forward LSTM + backward LSTM) to capture the dependency relationship between the past and the future; a fully connected layer to progressively compress features to a 1-dimensional output; a disaster output normalization layer to output the predicted disaster spread direction; and a disaster point loss function is constructed. A dynamic error-sensitive weighting mechanism is introduced, which dynamically adjusts the loss weights based on the magnitude of the prediction error. Priority is given to correcting large errors that could lead to serious consequences, while smaller errors within an acceptable range are given secondary attention. The correction error *l* for a single sample is defined as: Loss is calculated using the unit arc length to avoid prediction bias caused by boundary discontinuities; error classification weights are designed. , is represented as: ; E is the prediction error after boundary correction; It is the sensitivity coefficient; This is an enhancement coefficient that ensures larger errors are weighted more rapidly; through weight grading, the loss function is matched with the risk sensitivity of rescue decisions, and high-risk scenarios corresponding to large errors are prioritized for optimization; the final loss function is expressed as: ;in, It predicts the direction of disaster spread; H is the measured direction of disaster spread; H is the total number of samples; v is the sample index; a disaster output normalization layer is constructed, represented as: ;in, It is the normalized predicted direction of disaster spread; It is a modulo operation; the output is forcibly constrained to 0-360° to ensure that the prediction results conform to the physical meaning of the actual disaster spread direction;

[0047] High-precision point prediction provides a core basis for rescue decision-making, while boundary continuity processing avoids fatal misjudgments.

[0048] Example 4, see Figure 1 This embodiment is based on the above embodiment. In step S3, the quantification of the disaster spread range quantifies the uncertainty of the prediction, gives the possible range of the spread direction, provides fault tolerance space for rescue decisions, and avoids rescue errors due to the deviation of a single prediction value. The specific operation is as follows: based on the point prediction error, the error distribution is estimated and fitted, and a dynamic bandwidth of dynamic time scale is introduced to make the error distribution estimation adapt to different time scales. The error distribution is expressed as: ; Where G is the prediction error; ; This is an estimate of the probability density function of the prediction error, describing the likelihood of the error taking a certain value. If the prediction error is 5°... The highest value results in an error of 5°; N is the total number of error samples, n is the error sample index; h is the bandwidth, used to control the smoothness of the error distribution; It is the basic bandwidth. This is the time scale coefficient, where d is the number of days to forecast; This is the prediction error of the nth error sample; the prediction range obtained by combining the predicted disaster spread direction with the prediction error is processed across the boundary range, and the cross-boundary correction is expressed as: Where Inl is the processed cross-boundary range; and These are the original lower and upper bounds of the range, respectively; Used to describe the start and end relationships of the angle range; the confidence range is determined by the percentile method. At a 90% confidence level, the 5th and 95th percentiles of the error distribution are taken as the upper and lower boundaries, and one day is chosen as the time scale. The directions corresponding to the 5th and 95th percentiles of the error distribution are respectively... and Therefore, there is a 90% probability that the direction of the disaster's spread will fall within the next day. The warning results;

[0049] By quantifying uncertainty through range prediction, we can avoid over-reliance on a single forecast value in decision-making. In typhoon relief, we can clarify the possible fluctuation range of the disaster spread direction, expand the evacuation radius to leave sufficient safety redundancy, and ensure that the range conforms to the actual geographical direction through cross-boundary processing.

[0050] Example 5, see Figure 1 This embodiment is based on the above embodiment. In step S4, the prediction result evaluation involves dividing the disaster spread dataset into a test set and a training set, training the disaster spread prediction model based on the training set, and verifying the performance of the disaster spread prediction model based on the test set; a disaster spread prediction evaluation index is constructed, expressed as: ; Where M is the boundary correction mean error, measuring the overall bias of the prediction after boundary correction; R is the boundary correction mean square error; Q is the total number of test set samples, and q is the test set sample index; a range prediction evaluation index is constructed, including prediction range coverage and prediction range normalized mean width, and a coverage calibration term is introduced, expressed as follows: ; ; Where PP is the predicted coverage rate. These are the baseline calibration values. It is the calibration coefficient, and CL is the preset confidence level; It is the coverage indicator variable for the q-th sample, when the true direction falls within the predicted subrange after boundary correction. ,otherwise PW is the average width of the prediction range. PW is the total width of the predicted range for the q-th sample; the smaller the PW, the more accurate the warning; when the disaster spread prediction model converges to the training set loss, the disaster spread prediction model training is complete. Set a validation threshold, and verify the model performance based on the validation threshold. If the threshold is met, the disaster spread prediction model is established; otherwise, parameter optimization is performed. M, R, PP, and PW all have corresponding validation thresholds. For each sample in the test set, calculate the point prediction error to obtain M and R. For the range prediction result of each sample, determine whether the true value is covered by the range, calculate PP, and simultaneously calculate the width of each range to obtain PW. If M is lower than the M validation threshold and R is lower than the R validation threshold, the spread prediction is qualified. If PP is higher than the PP threshold and PW is lower than the PW threshold, the range prediction is qualified. When both the spread prediction and range prediction are qualified, the disaster spread prediction model is established.

[0051] By performing the above operations, this solution addresses the problems of boundary fracturing in directional prediction, the equalization deficiency in error processing, the static nature of uncertainty quantification, and risk mismatch in assessment indicators in general disaster reduction and early warning methods, which lead to poor accuracy in early warning. This solution addresses these issues by mapping angular errors to unit arc length through boundary continuity processing, combined with a disaster output normalization layer, to enforce constraints on the results and resolve boundary fracturing, preventing rescue teams from advancing towards the disaster source. Based on a dynamic error-sensitive weighting mechanism, it introduces graded error weights to achieve nonlinear increments in the weights of large errors. Furthermore, it introduces dynamic bandwidth error distribution estimation, combined with cross-boundary range correction, to ensure that the evacuation radius includes the entire risk area. Finally, it improves decision-making effectiveness based on the risk adaptability assessment of the prediction results, thereby enhancing the accuracy of early warning.

[0052] Example 6, see Figure 1 This embodiment is based on the above embodiment. In step S5, parameter optimization is to optimize the parameters of the disaster spread prediction model and minimize the disaster point loss function on the validation set. Specifically, it includes the following steps:

[0053] Step S51: Initial Solution Generation; Generate initial solutions using parameter combination mapping to ensure uniform initial distribution and coverage of the hyperparameter space; each solution represents a set of hyperparameter combinations, including the number of hidden units in the bidirectional LSTM layer, learning rate, temporal window size, regularization coefficient, batch size, base bandwidth, time scale coefficient, sensitivity coefficient, enhancement coefficient, and calibration coefficient; the generated initial solution is represented as: ; Where h is the position of the initial solution; and These are the lower and upper limits of the search space, respectively; and These are the mapping values ​​generated for the u-th and u+1-th iterations, respectively; These are control parameters;

[0054] Step S52: Fitness function design; for each solution, train a bidirectional LSTM and compute the fitness function on the validation set. The value is used as the fitness value; the smaller the value, the better the solution performance, that is, the better the combination of hyperparameters.

[0055] Step S53: Adjust the perturbation amplitude; dynamically adjust the perturbation amplitude according to the iteration process, strengthening global exploration in the early stage of iteration and strengthening local development in the later stage; represented as: And perform a mutation operation, represented as: ;in, It is a random perturbation vector, with each dimension following a uniform distribution [-1, 1] and being independent of each other; It is the amplitude of the disturbance; and These represent the minimum and maximum intensities, respectively; t is the current search iteration count, and maxt is the maximum search iteration count. and These are the positions of the i-th solution before and after mutation during the t-th search iteration, respectively.

[0056] Step S54: Inefficient replacement; Identify and replace inefficient solutions that have not been improved for a long time to avoid population diversity degradation and premature convergence; The inefficiency judgment rule is: if the fitness value improvement of a solution is less than a threshold after K consecutive iterations, it is marked as inefficient; Replace the inefficient solution with the generated initial solution.

[0057] Step S55: Solution Update; Design core control weights that dynamically change with iterations to adapt to the needs of different search stages, represented as: Location update is represented as: ;in, It is the core control weight; and These are the minimum and maximum weights, respectively. It is the updated position of the j-th dimension; It is the value of the j-th dimension of the optimal solution; yes The value of the j-th dimension;

[0058] Step S56: Solution evaluation; Set the maximum number of iterations. If the maximum number of iterations is reached or the fitness value of the optimal solution converges, then evaluate the performance of the disaster spread prediction model trained based on the optimal solution. If the standard is met, the disaster spread prediction model is established; otherwise, return to step S51.

[0059] By performing the above operations, this scheme addresses the problem that general disaster reduction and early warning methods suffer from static iterative processes for parameter optimization, degradation of population diversity, and consequently poor early warning effectiveness. It introduces phased control of perturbation amplitude to accelerate optimization convergence; dynamic balancing based on core control weights emphasizes preserving the diversity of the current solution; and replacement of inefficient solutions ensures the maintenance of population diversity in parameter combinations, thereby improving the effectiveness of early warning.

[0060] Example 7, see Figure 1 This embodiment is based on the above embodiment. In step S6, the disaster reduction and relief early warning is to collect disaster spread data in real time, and after preprocessing, input it into the disaster spread prediction model. The obtained disaster spread direction and error distribution are used to give early warning to the management personnel.

[0061] Example 8, see Figure 2 Based on the above embodiments, the disaster reduction and relief early warning system based on artificial intelligence provided by the present invention includes a data acquisition module, a disaster spread prediction module, a disaster spread range quantification module, a prediction result evaluation module, a parameter optimization module, and a disaster reduction and relief early warning module.

[0062] The data acquisition module obtains historical disaster spread data to obtain a disaster spread dataset;

[0063] The disaster spread prediction module is based on a disaster spread dataset and a bidirectional LSTM. It designs a disaster spread prediction model by constructing a loss function that includes a dynamic error-sensitive weighting mechanism.

[0064] The disaster spread range quantification module introduces a dynamic bandwidth fitting error distribution with a dynamic time scale to quantify the disaster spread prediction range and perform cross-boundary range processing.

[0065] The prediction result evaluation module evaluates the prediction performance using the boundary correction average error index to determine whether the disaster spread prediction model meets the standard.

[0066] The parameter optimization module optimizes the parameters of the disaster spread prediction model by dynamically adjusting the disturbance amplitude and core control weights.

[0067] The disaster reduction and relief early warning module is based on a disaster spread prediction model and provides early warning of disaster spread based on real-time collected disaster spread data.

[0068] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0069] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention.

[0070] The present invention and its embodiments have been described above. This description is not restrictive, and the accompanying drawings are only one embodiment of the present invention; the actual structure is not limited thereto. In conclusion, if those skilled in the art are inspired by this description and design similar structures and embodiments without departing from the spirit of the invention, such designs should fall within the protection scope of the present invention.

Claims

1. A disaster reduction and relief early warning method based on artificial intelligence, characterized in that: The method includes the following steps: Step S1: Data Acquisition; Obtain historical disaster spread data to obtain a disaster spread dataset; Step S2: Disaster spread prediction; Based on the disaster spread dataset and bidirectional LSTM, a disaster spread prediction model is designed by constructing a loss function that includes a dynamic error-sensitive weighting mechanism; Step S3: Quantify the disaster spread range; Introduce a dynamic bandwidth fitting error distribution with a dynamic time scale to quantify the disaster spread prediction range and perform cross-boundary range processing; Step S4: Evaluation of prediction results; The prediction performance is evaluated by using the boundary correction average error index to determine whether the disaster spread prediction model meets the standards; Step S5: Parameter optimization; The parameters of the disaster spread prediction model are optimized by dynamically adjusting the disturbance amplitude and core control weights; Step S6: Disaster Reduction and Relief Early Warning; Based on the disaster spread prediction model, provide early warning for disaster relief based on the real-time collected disaster spread data.

2. The disaster reduction and relief early warning method based on artificial intelligence according to claim 1, characterized in that: In step S2, the disaster spread prediction is based on a disaster spread dataset and a bidirectional LSTM to establish a disaster spread prediction model to predict the direction of disaster spread. The model structure includes: an input layer, a bidirectional LSTM layer, a fully connected layer, and a disaster output normalization layer; a disaster point loss function is constructed, a dynamic error-sensitive weighting mechanism is introduced, and the correction error of a single sample is defined; error classification weights are designed; the final loss function is obtained; and a disaster output normalization layer is constructed.

3. The disaster reduction and relief early warning method based on artificial intelligence according to claim 2, characterized in that: In step S3, the quantification of the disaster spread range is based on the error distribution estimation of the point prediction error, fitting the error distribution, and introducing the dynamic bandwidth of the dynamic time scale; the predicted range obtained by combining the predicted disaster spread direction with the prediction error is processed across the boundary range.

4. The disaster reduction and relief early warning method based on artificial intelligence according to claim 3, characterized in that: In step S4, the evaluation of the prediction results involves dividing the disaster spread dataset into a test set and a training set, training the disaster spread prediction model based on the training set, and verifying the performance of the disaster spread prediction model based on the test set; constructing disaster spread prediction evaluation indicators; constructing range prediction evaluation indicators, including prediction range coverage and prediction range normalized average width, and introducing a coverage calibration term; when the disaster spread prediction model converges to the loss on the training set, the disaster spread prediction model training is complete; setting a verification threshold, verifying the model performance based on the verification threshold; if the threshold is met, the disaster spread prediction model is established; otherwise, parameter optimization is performed.

5. The disaster reduction and relief early warning method based on artificial intelligence according to claim 4, characterized in that: In step S5, the parameter optimization involves optimizing the parameters of the disaster spread prediction model to minimize the loss function of the validation set, specifically including the following steps: Step S51: Initial solution generation; Initial solutions are generated using parameter combination mapping; Each solution represents a set of hyperparameter combinations; Step S52: Fitness function design; for each solution, train a bidirectional LSTM and compute the fitness function on the validation set. , as the fitness value; Step S53: Adjust the perturbation amplitude; dynamically adjust the perturbation amplitude according to the iteration process; and perform mutation operations; Step S54: Inefficient replacement; if the fitness value improvement of the solution after K consecutive iterations is less than the threshold, it is marked as inefficient; replace the inefficient solution with the generated initial solution. Step S55: Unlock and update; Step S56: Solution evaluation; Set the maximum number of iterations. If the maximum number of iterations is reached or the fitness value of the optimal solution converges, then evaluate the performance of the disaster spread prediction model trained based on the optimal solution. If the standard is met, the disaster spread prediction model is established; otherwise, return to step S51.

6. The disaster reduction and relief early warning method based on artificial intelligence according to claim 5, characterized in that: In step S5, the solution update involves designing the core control weights and updating their positions.

7. The disaster reduction and relief early warning method based on artificial intelligence according to claim 6, characterized in that: In step S1, the data acquisition involves obtaining historical disaster spread data; labeling the disaster spread direction; and performing preprocessing to obtain a disaster spread dataset.

8. The disaster reduction and relief early warning method based on artificial intelligence according to claim 7, characterized in that: In step S6, the disaster reduction and relief early warning involves collecting disaster spread data in real time, preprocessing it, and then inputting it into the disaster spread prediction model. The obtained disaster spread direction and error distribution are then used to provide early warnings to management personnel.

9. An artificial intelligence-based disaster reduction and relief early warning system, used to implement the artificial intelligence-based disaster reduction and relief early warning method as described in any one of claims 1-8, characterized in that: It includes a data acquisition module, a disaster spread prediction module, a disaster spread range quantification module, a prediction result evaluation module, a parameter optimization module, and a disaster reduction and relief early warning module; The data acquisition module obtains historical disaster spread data to obtain a disaster spread dataset; The disaster spread prediction module is based on a disaster spread dataset and a bidirectional LSTM. It designs a disaster spread prediction model by constructing a loss function that includes a dynamic error-sensitive weighting mechanism. The disaster spread range quantification module introduces a dynamic bandwidth fitting error distribution with a dynamic time scale to quantify the disaster spread prediction range and perform cross-boundary range processing. The prediction result evaluation module evaluates the prediction performance using the boundary correction average error index to determine whether the disaster spread prediction model meets the standard. The parameter optimization module optimizes the parameters of the disaster spread prediction model by dynamically adjusting the disturbance amplitude and core control weights. The disaster reduction and relief early warning module is based on a disaster spread prediction model and provides early warning of disaster spread based on real-time collected disaster spread data.

Images